|Home | About | Journals | Submit | Contact Us | Français|
Stem cell homing and breast cancer metastasis are orchestrated by the chemokine SDF-1 and its receptor CXCR4. Here, we report the nuclear magnetic resonance (NMR) structure of a constitutively dimeric SDF-1 in complex with a CXCR4 fragment that contains three sulfotyrosine residues important for a high-affinity ligand-receptor interaction. CXCR4 bridged the SDF-1 dimer interface so that sulfotyrosines sTyr7 and sTyr12 of CXCR4 occupied positively charged clefts on opposing chemokine subunits. Dimeric SDF-1 induced intracellular Ca2+ mobilization but had no chemotactic activity; instead, it prevented native SDF-1-induced chemotaxis, suggesting that it acted as a potent partial agonist. Our work elucidates the structural basis for sulfotyrosine recognition in the chemokine-receptor interaction and suggests a novel strategy for CXCR4-targeted drug development.
Chemokines direct homeostatic and proinflammatory immune responses by activating specific guanine nucleotide-binding protein (G protein)-coupled receptors (GPCRs) to induce cell migration along a gradient of increasing concentration of chemokine. The ~50 known chemokines share a conserved tertiary fold and are grouped into four subfamilies (C, CC, CXC, and CX3C) according to the spacing of conserved cysteines near the N-terminus. Many chemokine signaling pathways are also vital for cell migration in normal development or in abnormal conditions such as tumor metastasis. For example, the CXC chemokine stromal cell derived factor-1 (SDF-1, also known as CXCL12) and its receptor CXCR4 are essential for proper fetal development. Sdf1−/− or Cxcr4−/− mice die in utero due to defects in hematopoiesis, vascularization of the intestine, cerebellar formation, and heart development (1–3). CXCR4 is also the major coreceptor for T-tropic (X4) strains of human immunodeficiency virus 1 (HIV-1), and SDF-1 inhibits HIV-1 infection (4–7). Additionally, SDF-1 and CXCR4 mediate cancer cell migration and metastasis (8). Treatment with CXCR4-neutralizing antibodies reduces metastatic tumor formation in a mouse model of human breast cancer (8). CXCR4 is found in cells from over 20 types of cancer, which metastasize to tissues that secrete SDF-1, including the bone marrow, lung, liver, and lymph nodes (9).
Peptides derived from the N-terminal domains of chemokine receptors bind specifically to their respective chemokine ligands (10, 11). High-affinity binding to SDF-1 requires the extracellular N-terminal domain of CXCR4 (12), which must be posttranslationally modified by sulfation at three tyrosine residues (Tyr7, Tyr12, and Tyr21) (13, 14). Other chemokine receptors, including CCR5, CCR2B and CX3CR1, are similarly modified at one or more tyrosine residues (15–18). To define the basis for sulfotyrosine recognition in a chemokine-receptor signaling complex, we solved the structures of the extracellular N-terminal domain of CXCR4 in its unmodified, singly sulfated and fully sulfated forms when bound to its ligand SDF-1. Our nuclear magnetic resonance (NMR) studies revealed a symmetric 2:2 complex in which the binding of CXCR4 stabilized dimeric SDF-1 and each receptor sulfotyrosine occupied a unique site on the chemokine. Unexpectedly, the constitutively dimeric SDF-1 protein, which was employed for structural studies, blocked CXCR4-mediated chemotaxis at low nanomolar concentrations. These results provide the first view of sulfotyrosine recognition in a chemokine-receptor complex at atomic resolution and suggest a novel strategy for inhibition of CXCR4 signaling with oligomeric ligands.
We and others have shown that peptides corresponding to the N-terminus of CXCR4 bind to SDF-1 with micromolar affinity (13, 19), but attempts to solve the NMR structure of a complex containing SDF-1 and the N-terminus of CXCR4 were compromised by spectral broadening arising from the equilibrium between monomeric and dimeric forms of SDF-1. Because CXCR4 has been purified as a ligand-independent dimer (20) and binding of the N-terminal 38 residues of CXCR4 (p38) promotes SDF-1 dimerization (13), we engineered an SDF-1 protein to limit exchange between complexes of different stoichiometries. Guided by the crystal structure of SDF-1 (21), we identified Leu36 and Ala65 as residues at the dimer interface that could be replaced with Cys residues to form a pair of symmetric, intermolecular disulfide bonds (Fig. 1A). The SDF-1 double mutant [SDF1(L36C/A65C)] migrated as a stable dimer under nonreducing SDS-PAGE (Fig. 1B), and its translational self-diffusion coefficient, as measured by pulsed-field gradient NMR, was consistent with that of a dimeric species (22) (Fig. 1C). We confirmed the presence of disulfide bonds linking the two monomers and solved the structure of SDF1(L36C/A65C) by NMR. The structure of covalently-locked, symmetric SDF1(L36C/A65C) dimer (hereafter referred to as SDF12) was superimposable on that of a dimer of wild-type SDF-1, which was determined crystallographically (Fig. 1D). SDF12 displays the canonical chemokine fold in which a flexible N-terminus is connected by the N-loop to a three-stranded antiparallel β-sheet and a C-terminal α-helix.
Binding of p38 to15N-labeled wild-type SDF-1 induced chemical shift perturbations attributable to a combination of SDF-1 dimer formation and peptide binding (13). Gozansky et al. observed similar chemical shift patterns following the interaction of SDF-1 with the N-terminus of CXCR4, but incorrectly assumed that SDF-1 was purely monomeric (13, 19, 23). Hence, the CXCR4 N-terminal binding surface they identified incorrectly included the SDF-1 dimer interface (13, 19). Importantly, neither they nor we could solve a structure of native SDF-1 in complex with the N-terminus of CXCR4. Our titration of [U-15N]-p38 with SDF-1, which showed extreme NMR line broadening, explains why no structure could be obtained. The line-broadening resulted from p38 binding to SDF-1 that was fluctuating between its monomeric and dimeric states, thereby producing a weak NMR signal for the CXCR4 peptide and thwarting any chance of determining an NMR structure (13).
Because the locked dimer reduces the number of accessible states, interpretation of NMR spectra of SDF12 upon its binding to p38 was straightforward. Titration of 15N-labeled SDF12 with p38 (Fig. 2A) perturbed NMR signals of the residues of the N-loop but not those of the dimer interface (Fig. 2B), thus isolating signals for likely CXCR4:SDF-1 binding determinants. Because only one set of SDF12 signals was observed during the p38 titration and because addition of more than two molar equivalents of p38 induced no further chemical shift perturbations, we concluded that a 2:2 complex was formed by the binding of two p38 molecules to symmetric sites on the surface of SDF12 (Fig. 2C).
Tyrosine sulfation in the N-terminal domain of CXCR4 contributes substantially to the binding of SDF-1 (14). We showed previously that sulfation of Tyr21 enhances the affinity of p38 for SDF-1 by ~3-fold (13), and we observed that fully-sulfated p38-sY3 binds ~20-fold more tightly than the unsulfated peptide (apparent Kd = 0.2 ± 0.2 µM; data not shown). To understand the role of sulfotyrosines in CXCR4:SDF-1 binding, we solved the structures of unsulfated, selectively sulfated, and fully sulfated CXCR4 p38 peptides bound to SDF12. Recombinant [U-15N,13C]-labeled p38 was modified using purified tyrosyl protein sulfotransferase (13) to contain sulfotyrosine at position 21 (p38-sY1) or positions 7, 12, and 21 (p38-sY3) (Fig. 3A). Like the p38 and p38-sY1 peptides (13), free p38-sY3 displayed no secondary or tertiary structure in solution, and sulfation induced only local chemical shift changes.
For the structure of each complex, NOEs between SDF12 and the (sulfo)tyrosine side chains of CXCR4 (Fig. 3B) unambiguously defined the location of two p38 molecules on the chemokine. Each p38 peptide bound the chemokine in the same mode irrespective of the extent of sulfation (Fig. 3C–E). Two p38 molecules wrapped around the symmetric SDF12 dimer in an extended conformation that contained no secondary structure. Specific side chain–mediated contacts defined a path for the bound CXCR4 peptide that corresponded closely to the surface identified by 1H/15N chemical shift perturbations (Fig. 4A). In contrast, residues of the flexible N-terminus and the C-terminal α-helix of SDF12 were unperturbed by p38 binding, and the overall chemokine structure was unaffected.
CXCR4 stabilized SDF-1 by interacting with both subunits and recognizing unique features of the dimer interface. Near the CXCR4 N-terminus, each p38 peptide crossed the dimer interface such that sTyr7 and sTyr12 interacted with opposing SDF-1 subunits (Fig. 4A). In the membrane-proximal portion of the N-terminal domain of CXCR4, NOEs connected Pro27 to Gln59 in one subunit of SDF12 and to Leu66 in the opposing subunit, where the two C-terminal helices packed against each other. Structures of other sulfotyrosine-containing protein complexes show that the O-sulfonate group typically interacts with a positively charged side chain (24, 25). In a similar manner, each negatively-charged sulfotyrosine in CXCR4 occupied a unique positively-charged pocket on the SDF12 surface (Fig. 4B–D).
NOE constraints from Val23 in one subunit of SDF12 positioned the sTyr7 O-sulfonate to form a favorable electrostatic interaction with a positively-charged Arg20 side chain of SDF-1 (Fig. 4B). In a similar fashion, NOEs connected sTyr12 of p38 to Pro10 and Leu29 of the other subunit of SDF12 and placed the sulfotyrosine within ~3 Å of the positively-charged amino group of Lys27 (Fig. 4C). Residues connecting the N-terminal CXC motif with the β1 strand of SDF-1 (the “N loop”), particularly the RFFESH motif consisting of residues 12–17, were predicted from mutagenic studies to interact with the N-terminus of CXCR4 (12, 26, 27). We observed intermolecular NOEs between 1HN of Phe14 in SDF12 and the 1Hα of Gly19 from CXCR4 and from Val18 in the chemokine to sTyr21 of p38(Fig. 3B). NOEs also linked sTyr21 of p38 with Val49 in the β3 strand of SDF12 and positioned the sTyr21 O-sulfonate <5 Å from the guanidinium of Arg47 (Fig 4D), consistent with our earlier measurement of sulfotyrosine-specific chemical shift perturbations (13). Chemokine recognition of a receptor sulfotyrosine corresponding to sTyr21 by a basic pocket formed between the N-loop and the 40’s loop may be a common feature of the CXC family. Residues lining the sTyr21-binding pocket of SDF-1 (Val18, Arg47, and Val49) are conserved in at least half of the 16 CXC chemokines (28). A tyrosine corresponding to sulfotyrosine 21 of CXCR4 may likewise be found in all CXC family receptors except CXCR6. In contrast, neither sTyr7, sTyr12 nor their corresponding binding sites are conserved in the CXC receptors or chemokines.
To assess the relative contribution of each sulfotyrosine to SDF-1:CXCR4 binding, we designed a series of mutations of native SDF-1 to disrupt the putative binding sites individually and then measured Ca2+ mobilization in THP-1 cells, which express CXCR4 (29). We assessed the likely interaction between monomeric SDF-1 and the N-terminus of CXCR4 by looking at half of the SDF12:p38-sY3 structure (one SDF-1 subunit and one p38-sY3). Overall, substitutions in native SDF-1 that altered interactions observed in this model complex (Fig. 5A and B, red) resulted in higher EC50 values for CXCR4 activation as measured in Ca2+ mobilization assays, whereas substitutions that were not at the binding interface resulted in no change in EC50 (Fig. 5A and B, cyan). Table 1 lists the amino acid substitutions and Ca2+ mobilization EC50 values.
In the SDF12:p38-sY3 structure, the sTyr7 O-sulfonate formed a favorable electrostatic interaction with the positively-charged Arg20 in the chemokine. However, sTyr7 bound to one SDF-1 subunit whereas the majority of the rest of p38-sY3 bound to the other SDF-1 subunit. In the model of monomeric SDF-1 and CXCR4 peptide, the site of sTyr7 binding is not identified. It is clear that sTyr7 could not bind to Arg20 of monomeric SDF-1, and replacement of Arg20 with Ala in native SDF-1 produced no change in EC50 (Table 1). This suggests that if sTyr7 forms interactions with monomeric SDF-1, they are not through Arg20. At present, our structural studies cannot identify the location of sTyr7 binding to monomeric SDF-1, if such an interaction occurs.
In both the SDF12:p38-sY3 structure and the model, sTyr12 of p38-sY3 bound near to Lys27 of SDF-12. Substitutions of Ala and Glu acid at this position in SDF-1 increased the EC50 for Ca2+ mobilization to 10.1 and 16.8 nM, respectively. Val39 in the β2 strand of SDF-1 is directly across from Lys27 of the β1 strand and a Val39 → Ala39 substitution increased the EC50 to 27.1 nM. Also, the sTyr21 O-sulfonate is near the guanidinium of Arg47 (Fig 4D). An Arg47 → Ala47 substitution in native SDF-1 changed the EC50 to 14.1 nM, and replacement of the positively-charged Arg side chain with a negatively-charged Glu drastically altered activation (Arg47 → Glu47, EC50 = 654 nM) relative to wild-type SDF-1 (EC50 = 3.6 nM). Substitution of Val49, which has NOEs to sTyr21, with Ala also showed a 2.4-fold increase in EC50.
In human embryonic kidney (HEK) 293 cells, Tyr21 is sulfated to a higher degree than are Tyr7 and Tyr12 and sTyr21 contributes the most to SDF-1 binding to the expressed CXCR4 (14). The extent of sulfation of the Tyr residues of CXCR4 has not been characterized in THP-1 cells, but our results are consistent with those of Farzan et al. (14) because disruption of the SDF-1 binding site for sTyr21 had the greatest effect. The results from our mutagenesis studies are consistent with previous studies and suggest that the structure of SDF12 with the various CXCR4 peptides contributes to an understanding of the binding and activation of CXCR4 by native SDF-1.
Solving the NMR structure of the SDF12:CXCR4 complex required that the chemokine exist as a disulfide-stabilized dimer. To determine whether the constitutively dimeric chemokine retained biological activity, we compared Ca2+ mobilization and chemotactic responses induced by SDF-1 in THP-1 cells with those of SDF12. Robust activation of CXCR4 was observed for both wild-type SDF-1 (EC50 = 3.6 nM) and SDF12 (EC50 = 12.9 nM) in the Ca2+ mobilization assay (Fig. 6A). AMD3100, a small-molecule antagonist of CXCR4 (30), competed with both ligands with IC50 values of 3.3 nM for SDF-1 and 3.2 nM for SDF12 (data not shown). Unexpectedly, although 1 to 30 nM of wild-type SDF-1 induced chemotactic migration, the constitutively dimeric SDF12 failed to attract cells in a transwell chemotaxis assay at concentrations of up to 1 µM (Fig. 6B). Because SDF12 bound to CXCR4 and induced a Ca2+ mobilization response but exhibited no chemotactic activity, we speculated that it might block chemotaxis in response to wild-type SDF-1. Indeed, migration of THP-1 cells in response to 10 nM wild-type SDF-1 was potently inhibited by increasing concentrations of SDF12 (IC50 = 4 nM) (Fig. 6C).
In cell-based assays, chemokines typically induce chemotactic migration over a relatively narrow concentration range. Like other chemokines, SDF-1 exhibited a biphasic concentration dependence that decreased and ultimately ceased at higher concentrations (Fig. 6B). Because the locked SDF12 dimer inhibited chemotaxis (Fig. 6C), we speculated that low concentrations of monomeric SDF-1 might stimulate chemotaxis, whereas dimeric SDF-1, promoted by binding to heparin or CXCR4, might be present at higher concentrations and could therefore interfere with chemotactic signaling.
To test this hypothesis, we conducted chemotaxis assays in which we compared the responses of cells to an SDF-1 mutant that remains monomeric at higher concentrations to the responses of cells to wild-type SDF-1. If inactivation is indeed due to dimerization, SDF1(H25R), which has a dimer Kd ~10-fold higher than SDF-1 (22), should resist inactivation at higher concentrations and maintain a chemotactic response at concentrations at which the activity of SDF-1 decreases. Both proteins induced a dose-dependent chemotactic response from 1–30 nM and had similar EC50 values in Ca2+ mobilization assays, but SDF1(H25R) promoted cell migration much more strongly than did SDF-1 at higher concentrations (70–100 nM) before returning to baseline levels (Fig. 6D). Based on these results, we speculate that a shift in the oligomeric state of SDF-1 regulates chemotaxis, perhaps through a change in the kinetics of CXCR4 internalization.
Tyrosine sulfation has been predicted or observed for the N-terminal extracellular domain of most chemokine receptors (31). This post-translational modification contributes to high affinity binding of chemokine ligands and other binding partners such as the gp120 protein of HIV-1 (15). The structures of the SDF-1:CXCR4 complexes reported here provide the first illustration of how chemokines can recognize specific patterns of sulfotyrosine modification in their respective receptors. We validated these structural results in the context of the wild-type chemokine by performing functional assays on a panel of SDF-1 mutant proteins. Substitution of residues that interact with CXCR4 in the SDF12:p38-sY3 complex correlates strongly with changes in the EC50 for Ca2+ mobilization response in THP-1 cells. We previously reported that binding to the N-terminus of CXCR4 promotes SDF-1 dimer formation (13). It is now clear that the N-terminus of CXCR4 promotes dimerization of SDF-1 by contacting specific sulfotyrosine recognition sites on both sides of the dimer interface.
Although the functional role of chemokine dimers is not fully understood (32–37), dimerization is essential for the in vivo function of the CC chemokines monocyte chemoattractant protein 1 (MCP-1), RANTES [regulated upon activation, normal T cell-expressed and –secreted], and macrophage inflammatory protein 1β (MIP-1β) (38) and the CXC chemokine interferon-induced protein of 10 kD (IP-10, also known as CXCL10) (39). Structural differences between CC dimers and CXC dimers result in markedly different capacities for binding to GPCRs. The N-terminus of a CC chemokine participates directly in receptor activation (40), but also forms the dimer interface. Consequently, a disulfide-linked MIP-1β dimer fails to bind to its receptor CCR5 because critical binding determinants are buried in the dimer interface (37). In contrast, the N-terminus in a CXC chemokine dimer remains available for receptor interactions (41). A disulfide-linked dimeric form of the CXC chemokine interleukin-8 (IL-8) induces a Ca2+ mobilization response in neutrophils with an EC50 (1.5 nM) comparable to that of wild-type IL-8 (4.5 nM) (32). Thus, whereas CC chemokines seem to act on their receptors exclusively as monomers, monomers and dimers may both participate in CXC chemokine signaling. Other physiological binding partners, such as heparin, can promote chemokine dimer formation, as we showed for SDF-1 (22). Also, in the solved structure of SDF-1 with a heparin disaccharide, SDF-1 is present as a dimer (42). Because residues such as Lys27 of SDF-1 are involved in binding to both heparin and CXCR4, one function of the N-terminus of CXCR4 may be to displace heparin prior to receptor binding.
If SDF-1 dimer formation alters CXCR4 signaling, as our results indicate (Fig. 6B and 6D), is there also a role for CXCR4 receptor dimerization? Chemokine receptors and other GPCRs are widely proposed to exist and function as dimers (43–46), but their detection and characterization remain controversial (47, 48). Our results do not report directly on the oligomeric state of the receptor, but CXCR4 has been purified from cells as a homodimer (20) and the structure of SDF12:p38 (Fig. 4A) illustrates how binding to the CXCR4 N-terminus promotes the dimerization of SDF-1 (13). Residues in the flexible N-terminus of SDF-1 are responsible for CXCR4 activation and thus may correspond to small molecule agonists of other GPCRs, such as for the β2-adrenergic receptor (β2-AR) (26, 49). The spacing of the ligand-binding sites in the crystal structure of dimeric β2-AR (50, 51) matches the ~40 Å distance separating the N-termini of an SDF-1 dimer, which suggests that formation of a functional 2:2 SDF-1:CXCR4 complex might be plausible. To account for the observed inhibition of CXCR4-mediated chemotaxis by SDF12 (Fig. 6C), we propose a model in which monomeric SDF-1 activates the full complement of signaling pathways required for chemotaxis, but binding of the dimeric ligand produces a 2:2 chemokine:receptor complex that stimulates intracellular calcium signaling but prevents cell migration (Fig. 6E).
Our results reveal the first details of sulfotyrosine recognition by a chemokine, and provide a structural basis for the enhancement of chemokine binding affinity by this posttranslational modification. In addition, the structures of SDF12 explain why binding to the N-terminus of CXCR4 induces dimerization of SDF-1 (13). However, the SDF12:p38 structure also illustrates an unexpected mode of chemokine inhibition. As a full agonist, wild-type SDF-1 induces a Ca2+ mobilization response and chemotactic migration. In measurements of THP-1 cells, SDF12 is both a partial CXCR4 agonist, stimulating Ca2+ mobilization, and a selective antagonist that blocks chemotaxis. Additional experiments are required to demonstrate whether inhibition by ligand dimerization is a general feature of the CXC chemokine family, which could be exploited for therapeutic benefit.
Tyrosine sulfation of CXCR4 p38 peptides was performed as described elsewhere (13). Two samples were used for each structure determination: [U-15N,13C]-SDF12 with unlabeled p38 peptide, and [U-15N,13C]-p38 with unlabeled SDF12 using a 1:1.25 (monomer subunit) molar ratio of labeled to unlabeled components in each case. Standard NMR techniques were used for generating chemical shift assignments for 15N/13C-labeled SDF12, p38, sY1 p38 and sY1 p38 (52). 3-dimensional 15N-edited NOESY-HSQC, 13C-edited NOESY-HSQC, and 13C(aromatic)-edited NOESY-HSQC spectra (τmix = 80 ms) were used to generate distance constraints. Intermolecular distance constraints were obtained from a 3D F1-13C-filtered/F3-13C-edited NOESY-HSQC spectrum (τmix = 120 ms). Backbone dihedral angle constraints were obtained from 1Hα, 13Cα, 13Cβ, 13C’, and 15N secondary shifts using TALOS. Initial structures were calculated using the NOEASSIGN module of the torsion angle dynamics program CYANA followed by iterative manual refinement to eliminate constraint violations. X-PLOR was used for further refinement, in which physical force field terms and explicit water solvent molecules were added to the experimental constraints. Tables S1 to S4 list the statistics for Procheck-NMR validation of the final 20 conformers.
THP-1 cells, a monocytic leukemia cell line, were obtained from ATCC. Ca2+-dependent Fluo-3 emission was measured at 25°C using a PTI spectrofluorometer with an excitation wavelength of 505 nm and emission was detected at 525 nm. Immediately before measurement, an aliquot of cells was washed and resuspended and allowed to equilibrate at 25°C for five minutes in the cuvette. After establishing a baseline (~100 s), chemokine was added and the Ca2+ mobilization response was monitored for ~350 seconds. Total fluorescence intensity was measured after lysing cells with 1% Triton X-100, followed by the addition of 50 mM EDTA. Ca2+ mobilization signals are reported as the ratio of the chemokine-induced fluorescence intensity maximum and the fluorescence intensity after cell lysis. Chemotaxis was assayed using Transwells (5 µm pore; Costar, Cambridge, MA). THP-1 cells were washed with PBS and migration buffer (RPMI 1640 containing 2 mg/mL of bovine serum albumin). 5 × 105 cells in 100 µL were placed in the top well and migration buffer containing the indicated doses of chemokine was added to the bottom wells. Plates were incubated for 3 h at 37°C and 5% CO2. Transwell inserts were then removed and cells that had migrated into the lower chamber were counted using a hematocytometer. Assays were also performed with SDF-1 present in both the lower and upper chambers or with no SDF-1 in the lower chamber as controls to measure chemokinesis and basal migration, respectively. The chemotactic index is computed as the number of cells that migrated in response to chemokine divided by the number of cells counted in the absence of chemokine.
Table S1. Statistics for 20 SDF12 conformers.
Table S2. Statistics for 20 conformers of the SDF12:p38 complex.
Table S3. Statistics for 20 conformers of the SDF12:p38-sY1 complex.
Table S4. Statistics for 20 conformers of the SDF12:p38-sY3 complex.